Loading...
1/*
2 * LZO1X Compressor from MiniLZO
3 *
4 * Copyright (C) 1996-2005 Markus F.X.J. Oberhumer <markus@oberhumer.com>
5 *
6 * The full LZO package can be found at:
7 * http://www.oberhumer.com/opensource/lzo/
8 *
9 * Changed for kernel use by:
10 * Nitin Gupta <nitingupta910@gmail.com>
11 * Richard Purdie <rpurdie@openedhand.com>
12 */
13
14#include <linux/module.h>
15#include <linux/kernel.h>
16#include <linux/lzo.h>
17#include <asm/unaligned.h>
18#include "lzodefs.h"
19
20static noinline size_t
21_lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
22 unsigned char *out, size_t *out_len, void *wrkmem)
23{
24 const unsigned char * const in_end = in + in_len;
25 const unsigned char * const ip_end = in + in_len - M2_MAX_LEN - 5;
26 const unsigned char ** const dict = wrkmem;
27 const unsigned char *ip = in, *ii = ip;
28 const unsigned char *end, *m, *m_pos;
29 size_t m_off, m_len, dindex;
30 unsigned char *op = out;
31
32 ip += 4;
33
34 for (;;) {
35 dindex = ((size_t)(0x21 * DX3(ip, 5, 5, 6)) >> 5) & D_MASK;
36 m_pos = dict[dindex];
37
38 if (m_pos < in)
39 goto literal;
40
41 if (ip == m_pos || ((size_t)(ip - m_pos) > M4_MAX_OFFSET))
42 goto literal;
43
44 m_off = ip - m_pos;
45 if (m_off <= M2_MAX_OFFSET || m_pos[3] == ip[3])
46 goto try_match;
47
48 dindex = (dindex & (D_MASK & 0x7ff)) ^ (D_HIGH | 0x1f);
49 m_pos = dict[dindex];
50
51 if (m_pos < in)
52 goto literal;
53
54 if (ip == m_pos || ((size_t)(ip - m_pos) > M4_MAX_OFFSET))
55 goto literal;
56
57 m_off = ip - m_pos;
58 if (m_off <= M2_MAX_OFFSET || m_pos[3] == ip[3])
59 goto try_match;
60
61 goto literal;
62
63try_match:
64 if (get_unaligned((const unsigned short *)m_pos)
65 == get_unaligned((const unsigned short *)ip)) {
66 if (likely(m_pos[2] == ip[2]))
67 goto match;
68 }
69
70literal:
71 dict[dindex] = ip;
72 ++ip;
73 if (unlikely(ip >= ip_end))
74 break;
75 continue;
76
77match:
78 dict[dindex] = ip;
79 if (ip != ii) {
80 size_t t = ip - ii;
81
82 if (t <= 3) {
83 op[-2] |= t;
84 } else if (t <= 18) {
85 *op++ = (t - 3);
86 } else {
87 size_t tt = t - 18;
88
89 *op++ = 0;
90 while (tt > 255) {
91 tt -= 255;
92 *op++ = 0;
93 }
94 *op++ = tt;
95 }
96 do {
97 *op++ = *ii++;
98 } while (--t > 0);
99 }
100
101 ip += 3;
102 if (m_pos[3] != *ip++ || m_pos[4] != *ip++
103 || m_pos[5] != *ip++ || m_pos[6] != *ip++
104 || m_pos[7] != *ip++ || m_pos[8] != *ip++) {
105 --ip;
106 m_len = ip - ii;
107
108 if (m_off <= M2_MAX_OFFSET) {
109 m_off -= 1;
110 *op++ = (((m_len - 1) << 5)
111 | ((m_off & 7) << 2));
112 *op++ = (m_off >> 3);
113 } else if (m_off <= M3_MAX_OFFSET) {
114 m_off -= 1;
115 *op++ = (M3_MARKER | (m_len - 2));
116 goto m3_m4_offset;
117 } else {
118 m_off -= 0x4000;
119
120 *op++ = (M4_MARKER | ((m_off & 0x4000) >> 11)
121 | (m_len - 2));
122 goto m3_m4_offset;
123 }
124 } else {
125 end = in_end;
126 m = m_pos + M2_MAX_LEN + 1;
127
128 while (ip < end && *m == *ip) {
129 m++;
130 ip++;
131 }
132 m_len = ip - ii;
133
134 if (m_off <= M3_MAX_OFFSET) {
135 m_off -= 1;
136 if (m_len <= 33) {
137 *op++ = (M3_MARKER | (m_len - 2));
138 } else {
139 m_len -= 33;
140 *op++ = M3_MARKER | 0;
141 goto m3_m4_len;
142 }
143 } else {
144 m_off -= 0x4000;
145 if (m_len <= M4_MAX_LEN) {
146 *op++ = (M4_MARKER
147 | ((m_off & 0x4000) >> 11)
148 | (m_len - 2));
149 } else {
150 m_len -= M4_MAX_LEN;
151 *op++ = (M4_MARKER
152 | ((m_off & 0x4000) >> 11));
153m3_m4_len:
154 while (m_len > 255) {
155 m_len -= 255;
156 *op++ = 0;
157 }
158
159 *op++ = (m_len);
160 }
161 }
162m3_m4_offset:
163 *op++ = ((m_off & 63) << 2);
164 *op++ = (m_off >> 6);
165 }
166
167 ii = ip;
168 if (unlikely(ip >= ip_end))
169 break;
170 }
171
172 *out_len = op - out;
173 return in_end - ii;
174}
175
176int lzo1x_1_compress(const unsigned char *in, size_t in_len, unsigned char *out,
177 size_t *out_len, void *wrkmem)
178{
179 const unsigned char *ii;
180 unsigned char *op = out;
181 size_t t;
182
183 if (unlikely(in_len <= M2_MAX_LEN + 5)) {
184 t = in_len;
185 } else {
186 t = _lzo1x_1_do_compress(in, in_len, op, out_len, wrkmem);
187 op += *out_len;
188 }
189
190 if (t > 0) {
191 ii = in + in_len - t;
192
193 if (op == out && t <= 238) {
194 *op++ = (17 + t);
195 } else if (t <= 3) {
196 op[-2] |= t;
197 } else if (t <= 18) {
198 *op++ = (t - 3);
199 } else {
200 size_t tt = t - 18;
201
202 *op++ = 0;
203 while (tt > 255) {
204 tt -= 255;
205 *op++ = 0;
206 }
207
208 *op++ = tt;
209 }
210 do {
211 *op++ = *ii++;
212 } while (--t > 0);
213 }
214
215 *op++ = M4_MARKER | 1;
216 *op++ = 0;
217 *op++ = 0;
218
219 *out_len = op - out;
220 return LZO_E_OK;
221}
222EXPORT_SYMBOL_GPL(lzo1x_1_compress);
223
224MODULE_LICENSE("GPL");
225MODULE_DESCRIPTION("LZO1X-1 Compressor");
226
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * LZO1X Compressor from LZO
4 *
5 * Copyright (C) 1996-2012 Markus F.X.J. Oberhumer <markus@oberhumer.com>
6 *
7 * The full LZO package can be found at:
8 * http://www.oberhumer.com/opensource/lzo/
9 *
10 * Changed for Linux kernel use by:
11 * Nitin Gupta <nitingupta910@gmail.com>
12 * Richard Purdie <rpurdie@openedhand.com>
13 */
14
15#include <linux/module.h>
16#include <linux/kernel.h>
17#include <asm/unaligned.h>
18#include <linux/lzo.h>
19#include "lzodefs.h"
20
21static noinline size_t
22lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
23 unsigned char *out, size_t *out_len,
24 size_t ti, void *wrkmem, signed char *state_offset,
25 const unsigned char bitstream_version)
26{
27 const unsigned char *ip;
28 unsigned char *op;
29 const unsigned char * const in_end = in + in_len;
30 const unsigned char * const ip_end = in + in_len - 20;
31 const unsigned char *ii;
32 lzo_dict_t * const dict = (lzo_dict_t *) wrkmem;
33
34 op = out;
35 ip = in;
36 ii = ip;
37 ip += ti < 4 ? 4 - ti : 0;
38
39 for (;;) {
40 const unsigned char *m_pos = NULL;
41 size_t t, m_len, m_off;
42 u32 dv;
43 u32 run_length = 0;
44literal:
45 ip += 1 + ((ip - ii) >> 5);
46next:
47 if (unlikely(ip >= ip_end))
48 break;
49 dv = get_unaligned_le32(ip);
50
51 if (dv == 0 && bitstream_version) {
52 const unsigned char *ir = ip + 4;
53 const unsigned char *limit = ip_end
54 < (ip + MAX_ZERO_RUN_LENGTH + 1)
55 ? ip_end : ip + MAX_ZERO_RUN_LENGTH + 1;
56#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && \
57 defined(LZO_FAST_64BIT_MEMORY_ACCESS)
58 u64 dv64;
59
60 for (; (ir + 32) <= limit; ir += 32) {
61 dv64 = get_unaligned((u64 *)ir);
62 dv64 |= get_unaligned((u64 *)ir + 1);
63 dv64 |= get_unaligned((u64 *)ir + 2);
64 dv64 |= get_unaligned((u64 *)ir + 3);
65 if (dv64)
66 break;
67 }
68 for (; (ir + 8) <= limit; ir += 8) {
69 dv64 = get_unaligned((u64 *)ir);
70 if (dv64) {
71# if defined(__LITTLE_ENDIAN)
72 ir += __builtin_ctzll(dv64) >> 3;
73# elif defined(__BIG_ENDIAN)
74 ir += __builtin_clzll(dv64) >> 3;
75# else
76# error "missing endian definition"
77# endif
78 break;
79 }
80 }
81#else
82 while ((ir < (const unsigned char *)
83 ALIGN((uintptr_t)ir, 4)) &&
84 (ir < limit) && (*ir == 0))
85 ir++;
86 if (IS_ALIGNED((uintptr_t)ir, 4)) {
87 for (; (ir + 4) <= limit; ir += 4) {
88 dv = *((u32 *)ir);
89 if (dv) {
90# if defined(__LITTLE_ENDIAN)
91 ir += __builtin_ctz(dv) >> 3;
92# elif defined(__BIG_ENDIAN)
93 ir += __builtin_clz(dv) >> 3;
94# else
95# error "missing endian definition"
96# endif
97 break;
98 }
99 }
100 }
101#endif
102 while (likely(ir < limit) && unlikely(*ir == 0))
103 ir++;
104 run_length = ir - ip;
105 if (run_length > MAX_ZERO_RUN_LENGTH)
106 run_length = MAX_ZERO_RUN_LENGTH;
107 } else {
108 t = ((dv * 0x1824429d) >> (32 - D_BITS)) & D_MASK;
109 m_pos = in + dict[t];
110 dict[t] = (lzo_dict_t) (ip - in);
111 if (unlikely(dv != get_unaligned_le32(m_pos)))
112 goto literal;
113 }
114
115 ii -= ti;
116 ti = 0;
117 t = ip - ii;
118 if (t != 0) {
119 if (t <= 3) {
120 op[*state_offset] |= t;
121 COPY4(op, ii);
122 op += t;
123 } else if (t <= 16) {
124 *op++ = (t - 3);
125 COPY8(op, ii);
126 COPY8(op + 8, ii + 8);
127 op += t;
128 } else {
129 if (t <= 18) {
130 *op++ = (t - 3);
131 } else {
132 size_t tt = t - 18;
133 *op++ = 0;
134 while (unlikely(tt > 255)) {
135 tt -= 255;
136 *op++ = 0;
137 }
138 *op++ = tt;
139 }
140 do {
141 COPY8(op, ii);
142 COPY8(op + 8, ii + 8);
143 op += 16;
144 ii += 16;
145 t -= 16;
146 } while (t >= 16);
147 if (t > 0) do {
148 *op++ = *ii++;
149 } while (--t > 0);
150 }
151 }
152
153 if (unlikely(run_length)) {
154 ip += run_length;
155 run_length -= MIN_ZERO_RUN_LENGTH;
156 put_unaligned_le32((run_length << 21) | 0xfffc18
157 | (run_length & 0x7), op);
158 op += 4;
159 run_length = 0;
160 *state_offset = -3;
161 goto finished_writing_instruction;
162 }
163
164 m_len = 4;
165 {
166#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(LZO_USE_CTZ64)
167 u64 v;
168 v = get_unaligned((const u64 *) (ip + m_len)) ^
169 get_unaligned((const u64 *) (m_pos + m_len));
170 if (unlikely(v == 0)) {
171 do {
172 m_len += 8;
173 v = get_unaligned((const u64 *) (ip + m_len)) ^
174 get_unaligned((const u64 *) (m_pos + m_len));
175 if (unlikely(ip + m_len >= ip_end))
176 goto m_len_done;
177 } while (v == 0);
178 }
179# if defined(__LITTLE_ENDIAN)
180 m_len += (unsigned) __builtin_ctzll(v) / 8;
181# elif defined(__BIG_ENDIAN)
182 m_len += (unsigned) __builtin_clzll(v) / 8;
183# else
184# error "missing endian definition"
185# endif
186#elif defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(LZO_USE_CTZ32)
187 u32 v;
188 v = get_unaligned((const u32 *) (ip + m_len)) ^
189 get_unaligned((const u32 *) (m_pos + m_len));
190 if (unlikely(v == 0)) {
191 do {
192 m_len += 4;
193 v = get_unaligned((const u32 *) (ip + m_len)) ^
194 get_unaligned((const u32 *) (m_pos + m_len));
195 if (v != 0)
196 break;
197 m_len += 4;
198 v = get_unaligned((const u32 *) (ip + m_len)) ^
199 get_unaligned((const u32 *) (m_pos + m_len));
200 if (unlikely(ip + m_len >= ip_end))
201 goto m_len_done;
202 } while (v == 0);
203 }
204# if defined(__LITTLE_ENDIAN)
205 m_len += (unsigned) __builtin_ctz(v) / 8;
206# elif defined(__BIG_ENDIAN)
207 m_len += (unsigned) __builtin_clz(v) / 8;
208# else
209# error "missing endian definition"
210# endif
211#else
212 if (unlikely(ip[m_len] == m_pos[m_len])) {
213 do {
214 m_len += 1;
215 if (ip[m_len] != m_pos[m_len])
216 break;
217 m_len += 1;
218 if (ip[m_len] != m_pos[m_len])
219 break;
220 m_len += 1;
221 if (ip[m_len] != m_pos[m_len])
222 break;
223 m_len += 1;
224 if (ip[m_len] != m_pos[m_len])
225 break;
226 m_len += 1;
227 if (ip[m_len] != m_pos[m_len])
228 break;
229 m_len += 1;
230 if (ip[m_len] != m_pos[m_len])
231 break;
232 m_len += 1;
233 if (ip[m_len] != m_pos[m_len])
234 break;
235 m_len += 1;
236 if (unlikely(ip + m_len >= ip_end))
237 goto m_len_done;
238 } while (ip[m_len] == m_pos[m_len]);
239 }
240#endif
241 }
242m_len_done:
243
244 m_off = ip - m_pos;
245 ip += m_len;
246 if (m_len <= M2_MAX_LEN && m_off <= M2_MAX_OFFSET) {
247 m_off -= 1;
248 *op++ = (((m_len - 1) << 5) | ((m_off & 7) << 2));
249 *op++ = (m_off >> 3);
250 } else if (m_off <= M3_MAX_OFFSET) {
251 m_off -= 1;
252 if (m_len <= M3_MAX_LEN)
253 *op++ = (M3_MARKER | (m_len - 2));
254 else {
255 m_len -= M3_MAX_LEN;
256 *op++ = M3_MARKER | 0;
257 while (unlikely(m_len > 255)) {
258 m_len -= 255;
259 *op++ = 0;
260 }
261 *op++ = (m_len);
262 }
263 *op++ = (m_off << 2);
264 *op++ = (m_off >> 6);
265 } else {
266 m_off -= 0x4000;
267 if (m_len <= M4_MAX_LEN)
268 *op++ = (M4_MARKER | ((m_off >> 11) & 8)
269 | (m_len - 2));
270 else {
271 m_len -= M4_MAX_LEN;
272 *op++ = (M4_MARKER | ((m_off >> 11) & 8));
273 while (unlikely(m_len > 255)) {
274 m_len -= 255;
275 *op++ = 0;
276 }
277 *op++ = (m_len);
278 }
279 *op++ = (m_off << 2);
280 *op++ = (m_off >> 6);
281 }
282 *state_offset = -2;
283finished_writing_instruction:
284 ii = ip;
285 goto next;
286 }
287 *out_len = op - out;
288 return in_end - (ii - ti);
289}
290
291int lzogeneric1x_1_compress(const unsigned char *in, size_t in_len,
292 unsigned char *out, size_t *out_len,
293 void *wrkmem, const unsigned char bitstream_version)
294{
295 const unsigned char *ip = in;
296 unsigned char *op = out;
297 unsigned char *data_start;
298 size_t l = in_len;
299 size_t t = 0;
300 signed char state_offset = -2;
301 unsigned int m4_max_offset;
302
303 // LZO v0 will never write 17 as first byte (except for zero-length
304 // input), so this is used to version the bitstream
305 if (bitstream_version > 0) {
306 *op++ = 17;
307 *op++ = bitstream_version;
308 m4_max_offset = M4_MAX_OFFSET_V1;
309 } else {
310 m4_max_offset = M4_MAX_OFFSET_V0;
311 }
312
313 data_start = op;
314
315 while (l > 20) {
316 size_t ll = l <= (m4_max_offset + 1) ? l : (m4_max_offset + 1);
317 uintptr_t ll_end = (uintptr_t) ip + ll;
318 if ((ll_end + ((t + ll) >> 5)) <= ll_end)
319 break;
320 BUILD_BUG_ON(D_SIZE * sizeof(lzo_dict_t) > LZO1X_1_MEM_COMPRESS);
321 memset(wrkmem, 0, D_SIZE * sizeof(lzo_dict_t));
322 t = lzo1x_1_do_compress(ip, ll, op, out_len, t, wrkmem,
323 &state_offset, bitstream_version);
324 ip += ll;
325 op += *out_len;
326 l -= ll;
327 }
328 t += l;
329
330 if (t > 0) {
331 const unsigned char *ii = in + in_len - t;
332
333 if (op == data_start && t <= 238) {
334 *op++ = (17 + t);
335 } else if (t <= 3) {
336 op[state_offset] |= t;
337 } else if (t <= 18) {
338 *op++ = (t - 3);
339 } else {
340 size_t tt = t - 18;
341 *op++ = 0;
342 while (tt > 255) {
343 tt -= 255;
344 *op++ = 0;
345 }
346 *op++ = tt;
347 }
348 if (t >= 16) do {
349 COPY8(op, ii);
350 COPY8(op + 8, ii + 8);
351 op += 16;
352 ii += 16;
353 t -= 16;
354 } while (t >= 16);
355 if (t > 0) do {
356 *op++ = *ii++;
357 } while (--t > 0);
358 }
359
360 *op++ = M4_MARKER | 1;
361 *op++ = 0;
362 *op++ = 0;
363
364 *out_len = op - out;
365 return LZO_E_OK;
366}
367
368int lzo1x_1_compress(const unsigned char *in, size_t in_len,
369 unsigned char *out, size_t *out_len,
370 void *wrkmem)
371{
372 return lzogeneric1x_1_compress(in, in_len, out, out_len, wrkmem, 0);
373}
374
375int lzorle1x_1_compress(const unsigned char *in, size_t in_len,
376 unsigned char *out, size_t *out_len,
377 void *wrkmem)
378{
379 return lzogeneric1x_1_compress(in, in_len, out, out_len,
380 wrkmem, LZO_VERSION);
381}
382
383EXPORT_SYMBOL_GPL(lzo1x_1_compress);
384EXPORT_SYMBOL_GPL(lzorle1x_1_compress);
385
386MODULE_LICENSE("GPL");
387MODULE_DESCRIPTION("LZO1X-1 Compressor");